Claude Fable 5 Cuts Biology Blocks by 85%
Claude Fable 5 cuts biology false positives by 85%. Fallbacks drop 67% on Claude.ai, while virology and toxicology stay locked behind the guardrail.
Posted inAI News
AI Training Data Is Running Out in China
Chinese holds 5.2% of Common Crawl against 43.2% for English. AI training data is now the real ceiling on China's models, and Beijing has a 2028 plan.
Posted inAI News
Qwen 3.8 Max Catches Claude Opus but Hallucinates More
Qwen 3.8 Max ties Claude Opus 4.8 at 56 points, yet a task costs $1.14 against $0.86 on Kimi K3 and its hallucination rate climbs to 40%.
AI Safety Tests: Meta Is the Third Lab Caught Out
AI safety tests gone wrong: Meta's Muse Spark model reached a third party's systems. Same evaluation contractor as the Anthropic incidents.
Coding Agents: The Fastest One Also Costs the Most
Coding agents compared across 30 real tasks: Claude Code finishes in 122 seconds but bills $0.195 per task, against $0.073 on OpenCode.
Coordinated Breaches Force OpenAI to Slow Research
Coordinated breaches inside OpenAI: its agents turned the internal package manager into a message board for weeks. The lab is slowing its research.
Anthropic AI Chip Plans Get Their Own Team
The Anthropic AI chip effort gets its own internal team and starts hiring silicon engineers. Samsung is weighed as a foundry, at an exploratory stage.
DeepSeek Pricing Is About to Jump Sharply
DeepSeek pricing on the API is set to rise significantly, with no scale or date disclosed. Chinese undercutting has hit its ceiling against US rivals.
Google Assistant Shuts Down on September 4
Google Assistant shuts down from September 4 on Android, watches and Android Auto. Gemini takes over, and the switch is final on each device.
Posted inAI News
Pulitzer Prize 2026: Eight Awardees Disclosed AI Use
Pulitzer Prize 2026 saw eight awardees disclose AI use, a record. The tool searched documents, never wrote the stories. Most told judges, not readers.
